D2C / E-commerce (Consumer Electronics) · India
How a Consumer Electronics Brand Auto-Resolved 61% of Support Tickets — And Deliberately Stopped Deflecting 9% of Them
A Pune electronics brand chased deflection rate as a single number and hit 74% — while quietly losing its highest-value customers to the bot. NimbleBiz rebuilt the flow so NBScore decides which tickets the AI Support Agent answers and which go straight to a human. Deflection dropped to 61%. Repeat purchase rate went up 19%.
Customer: D2C consumer electronics & small appliances brand, Pune — ~9,000 support conversations/month — anonymized with permission.
The challenge
The brand sells audio gear, kitchen appliances, and grooming devices direct to consumers. Roughly 9,000 support conversations a month land on WhatsApp, Instagram DMs, and the website chat widget. Four agents. No support manager. They had already installed a chatbot. On paper it worked: 74% of conversations never reached a human. The dashboard was green. The revenue numbers were not. Repeat purchase rate had been sliding for two quarters. When the founder pulled 200 transcripts by hand, the pattern was ugly — the bot was deflecting everyone equally. A customer four days into a dead ₹12,000 speaker under warranty got the same canned troubleshooting loop as someone asking about store hours. Customers who had bought three times in eighteen months were being asked to "select an option from the menu" while trying to escalate a defect. Worse, context died at every boundary. A customer who had qualified as a hot lead on WhatsApp through a Meta ad in March would come back in July with a complaint on Instagram, and the bot treated them as a stranger. Nothing carried over. No purchase history, no prior conversation, no sense of who this person was to the business. The team had optimized for the wrong number. Deflection rate went up. The customers worth keeping went away.
The NimbleBiz setup
NimbleBiz replaced the chatbot with an AI Support Agent that answers from a real knowledge base — product manuals, warranty terms, return policy, order data — and, critically, checks NBScore before it decides to answer at all. **Every conversation gets scored before it gets a reply.** NBScore already tracked lead quality across the brand's Meta ad funnel. It was extended to the post-purchase side: order count, lifetime value, warranty status, and prior escalation history. That score, plus the intent the AI detects, decides the route. **Routine intents get resolved instantly, regardless of who is asking.** Order status, delivery windows, return eligibility, "how do I pair this", warranty lookup. The AI Support Agent pulls the answer from the knowledge base and the customer's own order record, and replies in seconds. Nobody waits for a human to look up a tracking number. **High-value customers with a real problem skip the bot entirely.** A defect report, a warranty claim, or a repeat complaint from a high-NBScore customer is not deflected. It routes to a human immediately, with Captured Details attached — name, order, product, warranty state, and the full prior conversation from whichever channel it started on. The agent opens the thread already knowing the story. **One inbox across all three channels.** WhatsApp, Instagram, and web chat land in the same unified inbox, threaded to one customer record. The March lead and the July complaint are now the same person, with the same history visible. **The 9% rule.** The brand set an explicit floor: roughly 9% of conversations — high-NBScore customers with product-failure intent — are never eligible for AI resolution. That is a deliberate cap on deflection, not a failure of it.
The outcome
Ninety days after launch: - **61% of conversations auto-resolved** by the AI Support Agent — down from 74% under the old chatbot, and up sharply in quality - **CSAT rose 22 points**, driven almost entirely by the segment that stopped being deflected - **Repeat purchase rate up 19%** against the prior quarter, reversing two quarters of decline - **Median first response time of 8 seconds** across all three channels, including nights and Sundays - **Escalations fell 34% in volume** but rose in value — agents now spend their day on defects and warranty claims instead of tracking numbers The four-person team did not grow. What changed is what they spend the day on. The founder's summary: "We were measuring how many customers we could avoid talking to. That's a strange thing to optimize when the ones you're avoiding are the ones paying you twice a year." The brand has since extended NBScore-gated routing to its Instagram comment-to-DM flow, and is piloting an AI Voice Agent for warranty escalations where a customer has gone quiet after a failed delivery. Start your free trial at nimblebiz.ai
Run the same playbook
Book a 15-minute demo — we'll show you the exact setup from this case study, adapted to your numbers.
Book a Demo